Exploring electrophysiological signal features for drought stress assessment in plants

Climate change is increasing environmental stress, particularly rising temperatures and water scarcity, in both natural and human-managed systems such as agroecosystems and urban environments. Traditional methods for monitoring plant health in human-managed systems remain limited, underscoring the need for novel approaches. This study explored the potential of plant electrophysiological signals (EPS) and derived statistical features for the early detection of drought stress. The two main objectives of this research were: i) to identify EPS features that are both ecologically relevant and statistically robust for detecting drought stress, and ii) to develop statistical models that integrate these features. EPS data was collected from two drought-stress experiments, one on tomato plants and one on apricot trees. Seventeen features from both time and frequency domains were selected and evaluated. Two models, a logistic and a machine learning classifier, were developed and compared using accuracy, precision, and recall metrics. In apricots, ten time-domain features (Frequency Centre, Generalized Hurst Exponent, Hjorth Complexity, Hjorth Mobility, Kurtosis, Root Mean Squared Frequency, Root Variance Frequency, Shape Factor, Skewness and Standard Deviation) showed significant differences between stressed and control groups. In tomatoes, four frequency-domain features (Frequency Centre, Root Variance Frequency, Root Mean Squared Frequency, and Power Law Distribution Exponent) were significantly different. Model accuracy was approximately 50% for apricots and 66% for tomatoes, insufficient for practical deployment but indicative of potential. This study illustrates the potential value of plant EPS data, its derived statistical features, and models for developing early drought stress detection indicators in both agroecosystems and urban forestry management contexts.